TITLE: Forecasting the Ring Current Index Dst in Real Time
AUTHORS: T. Paul O’Brien and Robert L. McPherron
AFFILIAITON: Institute of Geophysics and Planetary Physics,
405 Hilgard, UCLA, Los Angeles, CA 90095-1567
ABSTRACT
Three models for the evolution of the ring current index Dst are evaluated for real-time implementation.
Each model provides the time evolution of Dst in terms of solar wind parameters. Real-time data resources
employed by the models are Kyoto Quicklook Dst and ACE real-time solar wind parameters. The
implementation described provides a forecast time of about 1 hour due to propagation of the solar wind
from the ACE spacecraft to the Earth. The models are evaluated for simulated real-time data availability,
and several error measures are provided. One-hour averages were used in keeping with the standard Dst
index. By all measures used the model of O’Brien and McPherron (1999) performs best.
INTRODUCTION
When the interplanetary magnetic field (IMF) embedded in the solar wind opposes the Earth’s
intrinsic magnetic field, substantial transfer of energy into the terrestrial magnetosphere occurs. If this
condition persists for several hours, the entire magnetosphere becomes disturbed—a situation termed a
geomagnetic storm (Gonzalez et al., 1994). The primary measure of the intenstiy of a geomagnetic storm is
the strength of the ring current, as measured by Dst. When hot ions are injected into the inner
magnetosphere, the geometry of the geomagnetic field causes them to drift around the Earth forming a
westward ring current. The magnetic field of the ring current decreases the field at the surface of the earth,
and this depression is measured as Dst (Sugiura and Kamei, 1991). A typical storm includes a substantial
ring current that develops over a few hours and then recovers over several days (Kamide et al., 1997).
Forecasting the state of the ring current is a necessity for forecasting the magnetic field in the
magnetosphere (Baker, 1998). The Dst index represents the longest commonly used measure of the state of
the ring current, and therefore it is essential in such forecasting. We will discuss the implementation and
evaluation of three of the many models that forecast Dst in real-time.
In 1975 Burton et al. introduced a simple differential equation for the evolution of Dst in terms of
solar wind conditions:
dDst *
Dst *
 Q(t ) 
dt

(1)
The Q in this equation is the rate of energy injection into the ring current, and is generally given in terms of
VBs, the dawn-to-dusk component of the interplanetary electric field. In terms of the solar wind velocity
and the interplanetary magnetic field (IMF) in Geocentric Solar Magnetospheric (GSM) coordinates, VBs
is defined as:
 VB
VBs   z
 0
Bz  0
Bz  0
(2)
The decay time  is usually given a value from 3 to 20 hours. Dst* is a corrected version of Dst with the
contamination from magnetopause currents removed. The correction generally has the following form:
Dst *  Dst  b Pdyn  c
(3)
Pdyn is the solar wind dynamic pressure, the force exerted on the magnetosphere by the flowing solar wind.
Hereafter, we will use the values of the parameters Q, , b, and c to distinguish between three models of the
evolution of the Dst index.
OPERATIONAL CONSIDERATIONS
In order to turn a model of the above form into a real-time forecast, one must have an initial Dst
and upstream solar wind measurements thereafter. The initial Dst is obtained from the Kyoto World Data
Center Quicklook Dst (http://swdcdb.kugi.kyoto-u.ac.jp/dstdir/dst1/q/Dstqthism.html). This index is
available every 12 hours typically, and may be delayed as much as 24 hours. The upstream solar wind
measurements are obtained at the L1 point by the ACE spacecraft, and are made available in real time at
the Space Environment Center (SEC) (http://sec.noaa.gov/ace/ACErtsw_home.html). The solar wind data
are downloaded from SEC in real time, and propagated ballistically to the Earth. The propagation time
from L1 to Earth is typically 1 hour, which is the reason we have a forecast rather than a nowcast. Once the
solar wind data is propagated to the Earth, 1-hour averages are calculated for the velocity, number density,
and magnetic field components in GSM. From these 1-hour averages, we calculate VBs and P dyn. We are
then able to use a simple forward-difference to integrate the Dst index from the Quicklook value to a 1hour forecast:

Dst * (t ) 
Dst (t  t )  Dst (t )  Q(t ) 
t
 

*
*
(4)
Conversions between Dst* and Dst are performed according to Equation (3). We have simulated the realtime operation of this system for 3 different models of Dst, and we include below their performance.
THREE MODELS
The first model we tried (AK1) has , b, and c as constants, and Q is a function of VBs. It is
essentially the form of the original model proposed by Burton et al. (1975), but we have determined new
optimal coefficients based on the OMNI data set. The second model (UCB) is that of Fenrich and Luhmann
(1998), which has b and c as constants, but  depends on VBs, and Q is a function of both P dyn and VBs.
The third model (AK2) is that of O’Brien and McPherron (1999), which has b and c as constants, but  and
Q depend on VBs. These three models are given in detail in Table 1. It should be noted that each model is a
single-step, first-order differential equation that provides the evolution of Dst entirely in terms of
interplanetary conditions.
Another method of modeling the evolution of Dst is provided by Vassiliadis et al. (1999), who use
10-minute average data to fit parameters to a state-dependent second-order differential equation. They find
that the parameters of the ring current model appear to depend on phase of the storm (dDst/dt), Dst, and the
input VBs, in such a way that the ring current model switches between first order and second order
behavior at different times in the storm. They also find a wider range of recovery times between 25 and 2.5
hours, but these are not commensurate with the simple relation found by O’Brien and McPherron (1999).
This more complex model was not included in our study.
ERROR EVALUATION
In order to estimate each model’s operational performance, we simulated real-time data
availability conditions. The simulation period was May 14 through December 31, 1998. We assumed that
the Kyoto Quicklook Dst would be available at the second most recent Noon or Midnight, thereby causing
a Kyoto lag of 12 to 24 hours behind real-time. We used the actual real-time ACE data, which we had
archived locally, for the simulation period. About 10% of the period was covered by solar wind data gaps
of 1-hour or more. In Figure 1 we show the performance of all three models relative to the Kyoto
provisional Dst for November 4 (day 308) through November 22 (day 326) of 1998. It is clear that the AK2
model best reproduces the variations in the Dst index. From day 320 onward, there is a large ACE data gap,
and one can see the three models decay smoothly to zero over different time scales. These are moderately
intense disturbances, not necessarily representative of how these models perform on more severe
disturbances. In general, it seems that the AK2 model Dst lies somewhere in between the UCB and AK1
values, with AK1 overshooting the minimum Dst and UCB recovering too quickly. None of the three
models accurately reproduces the minimum Dst on day 310. Summary error statistics are provided in Table
1. It should be noted that AK2 outperforms both of the other models in RMS error for the entire simulation
and for those times when Dst was below –50 nT. The larger errors for Dst below –50 nT are a reflection of
both the difficulty in predicting large Dst and the uncertainty in the calculation of disturbed Dst.
The final error analysis we would like to show is the distribution of errors. Figure 2 shows the
distribution, in 5 nT bins, of the errors made by each model. The UCB model seems to frequently provide a
Dst value that is too high while the AK1 model seems to frequently provide a value that is too low. All
three models have error distributions which are roughly Gaussian, but only AK2 is zero-centered. The
Gaussian shape may be indicative of unpredictable randomness in the system. The fact that AK2 is zerocentered suggests that it may not accumulate large errors after many hours of evolution with solar wind
data alone. Again, AK2 outperforms the other two models.
CONCLUSIONS
We have demonstrated an ability to provide somewhat reliable real-time Dst forecasts of about 1
hour. We have compared three models based on the ideas of Burton et al. (1975), and we have found that
the AK2 model of O’Brien and McPherron (1999) outperforms the other two. This evaluation is based on a
simulation of real-time data availability. The data needed for real-time implementation are Kyoto
Quicklook Dst as an initial value and ACE real-time solar wind measurements from L1 for hourly
evolution. The AK2 model seems to mitigate the overshoot and rapid recovery problems of the AK1 and
UCB models.
ACKNOWLEDGEMENTS
We would like to thank the Kyoto World Data Center for making the Quicklook and Provisional
Dst available, the ACE spacecraft team and SEC for making the real-time solar wind data available, and the
NSSDC for making the OMNI dataset available. We would like to acknowledge useful discussions with
Janet Luhmann regarding the UCB real-time Dst model. This work was graciously supported by NSF grant
NSF ATM 96-13667.
REFERENCES
Baker, D.N. (1998) What is space weather? Advances in Space Research, 22, pp. 7-16.
Burton, R.K., R.L. McPherron, and C.T. Russell (1975) An empirical relationship between interplanetary
conditions and Dst. Journal of Geophysical Research 80, pp. 4204-4214.
Fenrich, F. R., and J. G. Luhmann (1998) Geomagnetic response to magnetic clouds of different polarity.
Geophysical Research Letters, 25, pp. 2999-3002.
Gonzalez, W.D., J.A. Joselyn, Y. Kamide, H.W. Kroehl, G. Rostoker, B.T. Tsurutani, and V.M. Vasyliunas
(1994) What is a geomagnetic storm? Journal of Geophysical Research, 99, pp. 5771-5792.
Kamide, Y., W. Baumjohann, I. Daglis, W.D. Gonzalez, M. Grande, J.A. Joselyn, R.L. McPherron, J.L.
Phillips, G. Reeves, G. Rostoker, A.S. Sharma, H. Singer, B.S. Tsurutani, and V.M. Vasyliunas (1997)
Current understanding of magnetic storms: Storm/substorm relationships. Journal of Geophysical Research,
103, pp. 17,705-17,728.
O’Brien, T. P., and R. L. McPherron (1999) An Empricial Phase-Space Analysis of Ring Current
Dynamics: Solar Wind Control of Injection and Decay. Journal of Geophysical Research, in press.
Sugiura, M., and T. Kamei (1991) Equatorial Dst index 1957-1986, ISGI Publications Office.
Vassiliadis, D., A. J. Klimas, and D. N. Baker (1999) Models of Dst geomagnetic activity and of its
coupling to solar wind parameters, Physics and Chemistry of the Earth, 24, pp. 107-112.
Table 1. Comparison of the three real-time models.
Q (nT/hour)
(VBs in mV/m, Pdyn in nPa)
AK1
UCB
(Fenrich and
Luhmann,
1998)

(hours)
17.0
 2.47VBs
0
VBs  05.

7.7 VBs  4

Q
1/ 3
 3 VBs  4
 4.32(VBs  05. ) Pdyn VBs  05.
b
(nT/nPa)
8.74
15.8
c
(nT)
11.5
20
RMSE
(all Dst)
19 nT
21 nT
RMSE
(Dst<-50nT)
38 nT
40 nT
AK2
(O’Brien and
McPherron,
1999)
 4.4(VBs  0.5)
 9.74 
2.4 exp 
 4.69  VBs 
7.26
11
16 nT
24 nT
Model
Figure 1. Comparison of models to Kyoto Dst for a disturbed period from November 4-22 (day 308-326) of
1998. Values were obtained using the real-time simulation described in the text.
50
0
nT
-50
Kyoto Dst
AK1
UCB
AK2
ACE Gap
-100
-150
-200
308
310
312
314
316
318
320
322
UT Decimal Day (1998)
324
326
Figure 2. Distribution of hourly errors for each model. The bin size on the abscissa is 5 nT. Error is defined
as Model-Dst.
0.16
Fraction of All Points
0.14
0.12
AK1
UCB
AK2
Bin Size:
5 nT
0.1
0.08
0.06
0.04
0.02
0
-50
-40
-30
-20
-10
0
10
Error (nT)
20
30
40
50